Improving handwritten digit recognition using hybrid feature selection algorithm

Purpose The amount of features in handwritten digit data is often very large due to the different aspects personal handwriting, leading high-dimensional data. Therefore, employment a feature selection algorithm becomes crucial for successful classification modeling, because inclusion irrelevant or redundant can mislead modeling algorithms, resulting overfitting and decrease efficiency. Design/methodology/approach minimum redundancy maximum relevance (mRMR) recursive elimination (RFE) are two frequently used algorithms. While mRMR capable identifying subset that highly relevant targeted variable, still carries weakness capturing along with algorithm. On other hand, RFE flawed by fact those selected not ranked importance, albeit effectively eliminate less important exclude features. Findings hybrid method was exemplified binary between digits “4” “9” “6” “8” from multiple dataset. result showed + support vector machine (SVMRFE) better than both sole (SVM) mRMR. Originality/value In view respective strength deficiency RFE, this study combined these methods an SVM as underlying classifier anticipating make excellent complement SVMRFE.